An Osaka Metropolitan University team ran more than 200,000 Google Street View images through machine learning and matched them against the sleep of 1,089 working adults.

Research on neighbourhoods and health usually runs into the same wall. Asking people about their surroundings produces data that is subjective and slow to collect; measuring surroundings objectively, by sending surveyors out to score streets, produces data that is reliable and far too expensive to gather at the scale of a city.

A team at Osaka Metropolitan University tried a third route. Writing in Building and Environment, they ran more than 200,000 Google Street View images through machine learning to characterise streetscapes across Tokyo, then matched those characteristics against the sleep of 1,089 working adults living on the lower floors of apartment buildings.

“This study demonstrates the potential for evaluating streetscape characteristics across large geographic areas,” said Professor Daisuke Matsushita of the Graduate School of Human Life and Ecology, who led the research, adding that the method allows street conditions to be assessed over wide areas at relatively low cost.

What the images were used for

The analysis worked at eye level rather than from above. Satellite imagery can measure how much green cover a neighbourhood has, but it cannot say what a person standing on the pavement sees, and the view from the street is what a resident actually experiences.

The machine learning converted the photographs into labelled maps, identifying buildings, trees, roads, sidewalks, vehicles and people. From those labels the team derived both objective measures — how much greenery, how enclosed the street — and perceptual ones, including how safe a street appears.

The restriction to lower-floor residents is a deliberate piece of study design. Someone on the twentieth floor does not experience the street outside in the way someone on the second floor does, in noise, in view, or in the sense of who is passing. Restricting the sample keeps the exposure meaningful.

Safety, greenery, enclosure

Three findings pointed the same way. A stronger sense of neighbourhood safety was linked to longer sleep, and the streetscape appeared to shape that rating. Greenery, such as leafy street trees, was associated with longer sleep. High enclosure — many tall buildings and few open spaces — was associated with both longer sleep and fewer insomnia symptoms.

The greenery result came with a mechanism attached: the researchers suggest trees may support sleep indirectly, by making a street feel safer, rather than through any direct effect.

That indirect route is worth separating from the usual explanations offered for green space and health, which tend to involve air quality, temperature or psychological restoration. Here the proposed pathway is perceptual. A street with mature trees reads as cared for, settled and overlooked, and a person who feels safe where they live sleeps longer.

Enclosure is the more surprising of the three, since a canyon of tall buildings is not most people’s picture of a restful neighbourhood. In a dense city it may work much as the trees do: enclosed streets are overlooked from many windows, carry less through traffic and feel contained rather than exposed.

The walkability result

The finding that cuts against expectation concerns walkability. Highly walkable streetscapes — those with more sidewalks and traffic signs — were associated with a lower sense of safety and shorter sleep duration.

That sits awkwardly with a large body of urban health research treating walkability as straightforwardly beneficial, and the explanation the researchers offer is about what walkability infrastructure indicates. Busier, more crowded walkable streets may feel less safe or relaxing. Sidewalks and traffic signs are built where there is traffic and pedestrian volume to manage, so the markers of walkability are also markers of activity.

There is a plausible reading in which both findings are right. Walkability may be good for the health outcomes it is usually measured against — physical activity, cardiovascular risk — while being bad for sleep, because the qualities that make a street useful to walk along during the day are the qualities that make it noisy and busy at night. A street can be a good street and a poor place to sleep beside.

It is also worth noting what “walkability” means when derived from street imagery. The measure here is built from visible infrastructure, which is a proxy for the concept rather than the concept itself, and proxies of this kind can pick up correlates the researcher did not intend.

What the design cannot show

The study is observational and cross-sectional, which means everything was measured at one point in time. It can establish that people living on certain kinds of street reported sleeping longer; it cannot establish that the street is why.

The obvious alternative explanation is sorting. People do not land in neighbourhoods at random. Income, age, family circumstances, job and shift pattern all influence where someone can afford to live and how well they sleep, and the leafier, more enclosed, less busy streets of Tokyo are not randomly distributed with respect to any of those. Whether the study’s analysis adjusted for such factors is not something the release says.

The release also reports no effect sizes or statistics of any kind — no minutes of sleep, no confidence intervals. The directions are given without magnitudes, so it is impossible to tell from the release whether these associations are large enough to matter in practice or merely detectable in a sample of a thousand.

The method as the product

Matsushita’s framing suggests the technique is at least as much the point as the specific associations.

“Based on the technique used in this study, cities could potentially measure perceived safety, beauty, liveliness, and enclosure as well as pollution and noise,” he said. “We hope that opening up this new perspective creates further possibilities for designing healthier neighbourhoods.”

That is a reasonable ambition, with a caveat built into the data source. Street View imagery is captured by a commercial operator on its own schedule, with uneven coverage and timestamps that may be years out of date, and it exists in detail for wealthy, car-accessible cities more than for anywhere else. A method that depends on it inherits those limits, which matters most for the places where the health consequences of bad streets are likely to be worst.

The work was supported by the Japan Society for the Promotion of Science and the Japan Science and Technology Agency. The authors declare no competing financial interests.

---

Notes

1. “Urban eye-level landscapes and sleep health: Cross-sectional evidence from a megacity using machine learning for objective and perceptual assessment.” Building and Environment, 2026. DOI: 10.1016/j.buildenv.2026.114711.
2. Osaka Metropolitan University. “The look and feel of your neighborhood may affect your sleep.” EurekAlert!, 3 August 2026. https://www.eurekalert.org/news-releases/1138325.

Photo by MART PRODUCTION on Pexels.com